Skip to content

Benchmark

WildBench: llama_reward.Information/Advice seeking.K=-1

People compare model answers and choose which they prefer. Higher scores are better.

Models measured
32
Scores recorded
32
Sources
1
IQ weight
1.67

Score distribution

17.5
Mean
26.8
Median
28.6
Std dev
-16.8
10th percentile
48.5
90th percentile
-68.1
Min
-68.154.9
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.67 for chat models, measured over 32 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1GPT 4oallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data54.9
2Claude 3.5 Sonnetallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data51.8
3GPT-4 Turboallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data50.1
4Gemini 1.5 Proallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data48.5
5GPT-4-0125-previewallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data46.8
6Llama 3 70B Instructallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data43
7Reka Coreallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data40.2
8Yi 1.5 34B Chatallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data40.1
9Gemini 1.5 Flashallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data39.8
10Nemotron-4 340B Instructallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data38.9
11Claude 3 Opusallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data37.4
12yi-largeallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data37.1
13DeepSeek Coder V2 Instructallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data27.8
14GLM-4-9b-chatallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data27.7
15Qwen2 72B Instructallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data27.5
16Command R Plusallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data26.8
17Claude 3 Sonnetallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data22.1
18Claude 3 Haikuallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data15.3
19Starling LM 7B Betaallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data14.4
20Command R+allenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data12.3
21Llama 3 8B Instructallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data11.6
22Reka Flashallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data9.2
23Mistral Largeallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf, 2402· WildBench direct data7.4
24Mixtral 8x7B Instruct v0.1allenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data4.4
25Mistral 7B Instruct v0.2allenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data-0.525
26Reka Edgeallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data-3.7
27Nous Hermes 2 Mixtral 8x7B DPOallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data-6.4
28Tulu 2 DPO 70Ballenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data-9.2
29GPT-3.5 Turboallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data-16.8
30Llama-2-7b-chat-hfallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data-20.3
31Gemma 7B ITallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data-50
32Gemma 2B ITallenai/WildBench, llama_reward.Information/Advice seeking.K=-1, Llama-2-70b-chat-hf· WildBench direct data-68.1

Data credits: WildBench, Allen Institute for AI